DOS Attack Detection Using Fisher Score Based Feature Selection and Gated Recurrent Network in Computer Network

Abstract

Security threats is determine to be a significant issue caused in computer network due to malicious behaviour. Prevailing of these abnormal activities is due to injection of various cyber-attack by the attackers. A solution for this problem is to introduce Artificial Intelligence (AI) system for achieving effective DOS attack detection. Machine learning algorithm was used initially to detect DOS attacks. But low detection rare is a major drawback faced on using this algorithm. To overcome this issue, deep learning based DOS attack detection model is developed using Gated Recurrent Network (GRU). In this proposed model, initially the information regarding data storing and accessing by the users is obtained. The acquired data is pre-processed using mini-max normalization. Then, from the pre-processed data the features necessary for detection using selected using filter based technique namely fisher score. These selected features are given as input into GRU to detect DOS attack and non-attack data. The non-attack data is further encoded using Modified Random Linear Network Coding (MRLNC) and finally the encoded data using MRLNC is stored in cloud. Whereas in case of DOS attack data an alert message is provided to the user regarding the attack. Proposed deep learning based DOS detection model is implemented to evaluate its performance. Some of the statistical metrics such as accuracy, error, precision and sensitivity attained for proposed DOS attack detection model is 93%, 7%, 94% and 92%. This evaluation shows that improved DOS attack detection is achieved using proposed GRU model.